An Introduction to Longitudinal Research

Elisabetta Ruspini

• Publisher: Routledge
• Year: 2003
• ISBN:
  9781134510610
• Edition: 1
• Topic: Social Sciences
• Subtopic: Social Science Research & Methodology
• Index: Social Sciences

One of the major changes in the social science research landscape in recent years has been the introduction of computerised panel surveys in Europe and the US which make longitudinal data widely available to graduate students for the first time. Elisabetta Ruspini here provides a concise yet comprehensive introduction to the issues involved in this kind of research. This book:
* Defines the concept of longitudinal research
* Gives guidance on sources of longitudinal data in Europe and the US and their strengths and weaknesses
* Discusses the choices that need to be made in this kind of research - for instance the advantages and disadvantages of certain types of research data and of different types of analysis
* Highlights some of the problems involved, e.g. the issue of comparability within longitudinal research

Part I
Longitudinal research

1 What is longitudinal research?

The term ‘longitudinal’ will be used here to describe what can be defined as the minimum common denominator of a family of those methods which tell us about change at the individual micro level (Zazzo, 1967; Menard, 1991). This family is the opposite of that described by the term ‘cross-sectional research’.

‘Longitudinal’ is a rather imprecise term. Longitudinal data can be defined as data gathered during the observation of subjects on a number of variables over time. This definition implies the notion of repeated measurements (van der Kamp and Bijleveld, 1998). Basically, longitudinal data present information about what happened to a set of units (people, households, firms, etc.) across time. The participants in a typical longitudinal study are asked to provide information about their behaviour and attitudes regarding the issues of interest on a number of separate occasions in time (called the ‘waves’ of the study) (Taris, 2000). In contrast, cross-sectional data refer to the circumstances of respondents at one particular point in time (I shall expand on these points later). Thus, the term ‘longitudinal’ refers to a particular type of relationship between phenomena: the type which evolves over the course of time and is termed diachronic, the opposite of synchronic.

There are many different methods that can be used to collect longitudinal data, which means there are also many different types of research (Buck et al., 1994; Davies and Dale, 1994; Bijleveld et al., 1998; Ruspini, 1999, 2000a; Taris, 2000).

The most commonly used longitudinal designs are:

• repeated cross-sectional studies (trend), carried out regularly, each time using a largely different sample or a completely new sample;
• prospective longitudinal studies (panel), that repeatedly interview the same subjects over a period of time;
• retrospective longitudinal studies (event history or duration data) in which interviewees are asked to remember, and reconstruct, events and aspects of their own life-courses.

Of these three, prospective studies are considered the most ‘truly longitudinal’ (consequently preferable when analysing microsocial change), because they, periodically, gather information about the same individuals (Janson, 1990; Magnusson et al., 1991), who are asked the same sequence of questions at regular intervals. In particular, prospective longitudinal surveys provide the most reliable data on change in knowledge or attitudes, because longitudinal measures are collected while the subjective states actually exist. Indeed, some consider retrospective surveys to be ‘quasi-longitudinal’, both because they offer only an incomplete contribution to the study of causal processes and, above all, because of distortions due to inaccuracies in memories (Hakim, 1987: 97; Draper and Marcos, 1990; Dex, 1991; Taris, 2000). Each longitudinal design will be examined separately here (see Chapter 2 for details).

A cross-sectional survey studies a cross-section of the population at a specific moment or point in time. Here, the term ‘cross-section’ indicates a wide sample of people of different ages, education, religion and so on. Repeated crosssectional studies, such as the General Household Survey or the Family Expenditure Survey in Great Britain, the European Community Eurobarometer Surveys, the Italian National Institute of Statistics (ISTAT) Multipurpose Survey of Italian Families (Indagine Multiscopo sulle famiglie italiane) and the Bank of Italy Survey of Household Income and Wealth, can help in the study of social change. However, because these surveys are not based on the same sample, they only offer a means for analysing net changes at the aggregate level – the net effect of all the changes – (Firebaugh, 1997): e.g. a comparison between the incidence of poverty and the characteristics of the population below the poverty line at time t and at time t–1 or between the pool of employed and unemployed in two different years. Thus, cross-sections can tell us about populations either at one or at a series of points in time.

Longitudinal data tell us about change at the individual or micro level providing estimates of both net and gross change – that is, the analysis of flows between states – and other components of individual change (i.e. to disaggregate net change) (Rose, 2000: 27). Prospective longitudinal studies, especially Household Panel Studies (HPS), follow individuals and families over time by, periodically, re-interviewing the same subjects and providing multiple observations on each individual/household in the sample. Such studies involve not only a random sample of households, but also all those members and subsequent co-residents, partners and descendants who are repeatedly re-interviewed. Thus, these studies accumulate records of employment, income, family status and attitudes over extended periods. This makes it possible to study change at the individual, i.e. the micro, level (Hakim 1987; Rose and Sullivan, 1996; Gershuny, 1998, 2000), that is, to analyse changes within the institutional, cultural and social environments that surround the individual and shape the course of his/her life. Thus, they offer a basis for further study of the dynamics of social phenomena – an advantage Paul F. Lazarsfeld must have recognised when, in the late 1930s, he was the first to use longitudinal data when analysing the relation between radio advertisement and product sales/changes in public opinion. He suggested repeatedly interviewing the same respondents would clarify whether the radio advertisement was the cause or the effect of buying the product (Lazarsfeld, 1940). Thus, for Lazarsfeld the panel technique seemed to be one of the most promising for the future of a fuller understanding of human behaviour (Lazarsfeld, 1948).

Among the areas of panel research which have been identified as being of particular concern to policy-makers are the following (Rose, 2000):

• dynamic analyses of labour income (Joshi and Davies, 2000);
• analysis of career trajectories (Scherer 2000; Gallie and Paugam, 2000);
• poverty and income dynamics (Walker and Ashworth, 1994; Ashworth et al., 2000; Jarvis and Jenkins, 2000; Muffels 2000);
• the gender dimension of poverty (Ruspini, 2000b);
• child poverty, child achievement and parenting (Ashworth et al., 1992a, 1992b; Hill and Jenkins, 1999);
• well-being of the elderly (Coe, 1988; Burkhauser and Duncan, 1988, 1991; Bound et al., 1991; Lillard and Waite, 1995);
• social exclusion (Walker, 1995);
• analysis of welfare use (Walker and Ashworth, 1994);
• analysis of the achievements and failures of welfare states (Goodin et al., 1999);
• household change: household formation and dissolution (Blossfeld, 1995; Jarvis and Jenkins, 1998; Ermisch, 2000);
• dynamic issues of disability¹ (Adler, 1992; Eustis et al., 1995);
• transitions, e.g. into/out of the labour force; from youth to adulthood.

Event history or duration data offers a record of the events that have punctuated the life-course of a group of subjects. These concepts need to be clarified. Life-course is used to refer to the history of each family or individual and to the way this history evolves and changes over time (Saraceno, 1986). The life-course is determined by interdependent trajectories and transitions that subjects (individual or collective – woman, man, couple, firm) undergo during the course of their existence. Trajectories refer to the path taken, as time goes on, within a specific, relatively long-term experience or position – the family, work, etc. – one which often may continue for a large part of the individual’s lifespan. Transitions are fluctuations/changes within a trajectory: in other words, trajectories are characterised by the transitions, or changes, of social, economic and demographic interests which evolve in response to specific events (Elder, 1985). In this instance, ‘event’ is taken to mean a change, or a transition, from one discrete state to another, a passage which takes place at a specific point in time and which constitutes a radical departure from what came before the ‘catalysing’ event: e.g. marriage, the birth of a child, starting work, divorce, etc. (Allison, 1984). Thus, an event can be defined as a change that gives an individual new status, which differs from the previous status the individual had before the change took place.

This definition of an event enables us to visualise events as transitions between states (Rajulton, 1999). The most important transitions (e.g. the transition to adulthood) usually introduce a multiplicity of changes into individuals’ lives (Billari, 1998). However, apparently similar transitions may assume a different significance depending on the point at which they take place within a particular trajectory: going to university straight from school or after taking a few years out to work; having a child at 20 or at 40; being made redundant when a young adult and losing a job when middleaged with adolescent children to support (Olagnero and Saraceno, 1993). Thus, life-course dynamics arise from the interplay of trajectories and transitions, an interdependence played out over time and in relation to others (Elder, 1985).

Duration data are usually gathered using retrospective cross-sectional studies in which respondents are asked to remember events and aspects of their own life-courses. Typically, this is done domain by domain, beginning with the current situation and taking respondents backwards in time. In panel surveys, data may be collected at the first wave either retrospectively for a fixed initial reference period or as far back as a specific event, such as marriage or first employment (Skinner, 2000). While this design is both simple and cheap, these data are typically more complicated than those obtained with trend or panel techniques because detailed information is given for each episode – that is, a time span a unit of analysis (e.g. a woman/man) spends in a specific state – details about the duration and frequency of the event and about any other aspects which show marked diachronic variation. However, retrospective surveys do have clear limitations, both in the necessarily simplified form in which they are forced to reconstruct experiences and, above all, because memory often distorts reality when trying to recall past events (Dex, 1991). Hence, retrospective surveys are usually limited to significant but infrequent life events such as births, marriages, divorces and job changes (Rose, 2000: 12).

Research is rarely based on one investigative method alone; indeed, longitudinal research is commonly based on a mix of methods.² Some examples of longitudinal mixed designs are:

1 Repeated cross-sectional studies one part of which are done in the form of panel studies. For example, the British Social Attitudes Survey (BSA) or the Bank of Italy Survey of Household Income and Wealth (SHIW) are repeated regularly on a largely different sample but with a small part as a panel study (Jowell et al., 1992).

---

Example 1.1 Examples of Repeated Cross-Sectional Studies with a Small Panel section

The BSA is an annual survey that measures changes in social attitudes, with particular reference to influence upon the ways that people vote, which has been charting changing values in Britain since 1983. Core-funded by the Sainsbury Family Charitable Trust, its findings are based on hour-long interviews with a sample of 3,600 people. The survey is designed to yield a representative sample of adults aged 18 and over in England, Scotland and Wales. Since 1993 the sampling frame for the survey has been the Postcode Address File (PAF), a list of addresses (or postal delivery points) compiled by the Post Office. The sample is confined to those living in private households. People living in institutions are excluded, as are households whose addresses are not on the PAF. In most years three versions of the BSA questionnaire are fielded. Each ‘module’ of questions is asked either of the full sample (around 3,600 respondents) or of a random two-thirds or one-third of the sample. Two of the main purposes of the BSA series is to allow monitoring of patterns of continuity and change, and the examination of the relative rates at which attitudes change over time with respect to social issues. The subjects covered by the surveys are wideranging, but include housing and home ownership, work and unemployment, health and social care, education, business and industry, social security and dependency, tax and spending, the welfare state, transport, environment and the countryside, constitutional reform, law and order, civil liberties, moral issues and sexual mores, racism and sexism, social inequality, religion, politics and governance.

Web sites:http://www.qb.soc.surrey.ac.uk/surveys/bsa/bsaintro.htm

http://www.data-archive.ac.uk/findingData/bsaAbstract.asp

The SHIW was launched in 1965. Twenty-three further surveys have been conducted since then, yearly until 1987 (except for 1985) and every two years thereafter. The aim of the survey is to gather information about the economic behaviour of Italian families at the microeconomic level. Data on family income, saving, expenditure, consumer durables and real wealth have been collected since 1966, while the acquisition of details concerning total consumption expenditure started in 1980. The basic survey unit is the household, which is defined in terms of family relationships, that is, as a group of individuals linked by ties of blood, marriage or affection, sharing the same dwelling and pooling all or part of their incomes. Persons living in nursing homes for the aged or ill, in prisons or military installations are not included. The survey has a panel section, corresponding to: 15.0 per cent of the households between 1987 and 1989; 26.7 per cent between 1989 and 1991; 42.9 per cent between 1991 and 1993; 44.8 per cent between 1993 and 1995, 37.3 per cent between 1995 and 1998 and 48.4 per cent between 1998 and 2000 (Brandolini and Cannari, 1994; D’Alessio and Faiella, 2002).

Web site: http://www.bancaditalia.it/

---

2 Prospective studies that gather information systematically through the use of calendars and/or suitable batteries of questions which aim to retrospectively investigate the life of the interviewee but not necessarily enquire about the same subject each time. One typical example are Household Panel Studies (HPSs), the most important of these being the Panel Study of Income Dynamics (PSID) in the United States, the German Socio-Economic Panel (GSOEP) in Germany and the British Household Panel Study (BHPS) in Britain.

---

Example 1.2 Household Panel Studies

The PSID is the longest running household panel today. It is a prospective longitudinal study, set up in 1968 in the Survey Research Center – Institute for Social Research (University of Michigan) and based on a proportional sample of the resident population of the United States (men, women and children) and their families. Since 1985, the PSID has also been collecting detailed, retrospective data on the histories, both family and matrimonial, of the subjects in the sample (the Demographic History Files).

Web site: http://www.isr.umich.edu/src/psid/

The GSOEP is a representative longitudinal study of private households in the Federal Republic of Germany. It has been modelled on the PSID. Its first wave went into the field in 1984, with a sample of 5,921 households and 12,245 individuals. The same private households, persons and families have been surveyed annually since 1984. The GSOEP has been developed and is carried out by the Project Group ‘Socio-Economic Panel’ at the German Institute for Economic Research (DIW), Berlin. In co-operation with the DIW, the Centre for Policy Research at Syracuse University has prepared an English language publicuse version of the GSOEP for use by the international research community. The public-use version of the GSOEP is offered to researchers throughout the world for use when studying the socio-economic characteristics of persons living in Germany (Butrica, 1996a). In order to reduce the risk of identifying individuals or households, this file does not include detailed information on nationality or region and represents a 95 per cent random sample of the original data. GSOEP data cover a wide range of subjects including: household composition; occupational and family biographies; employment and professional mobility; earnings; health; personal satisfaction as well as subjects covered in topical modules of the survey. Topical modules add questions on a variety of topics not covered in the core section: social security; education and training; allocation of time; family and social services. Moreover, two calendars are included in the core questionnaires: these calendars record monthly retrospective information on labour force participation and income.

Web sites: http://www.diw.de/english/sop/index.html

http://www.diw.de/english/sop/uebersicht/

The first three waves of the BHPS gathered information about employment histories, both family and matrimonial histories, and on individual demographic behaviour. The research group which conducts the BHPS has recently created a file, the BHPS Work-Life History Project, which puts together prospective and retrospective data (gathered during the second wave) concerning the employment conditions and the work/employment histories of those interviewed. This has made it possible to trace and reconstruct the occupational biographies of interviewees from the moment they entered the labour market up to the time of the most recent wave. More precisely, the Work-Life History Project is based on all sources of employment status and occupational information in the BHPS. These files combine information from:

• the inter-wave job history, all waves;
• the main file for current individual status, all waves;
• retrospective occupational history, wave 3;
• retrospective employment-status history, wave 2.

Web site: :http://www.data-archive.ac.uk/doc/3954%5Cmrdoc%5Cpdf%5Cnewman.pdf

---

3 Cohort studies that are also prospective and/or retrospective (two British examples of this being the National Child Development Study and the Birth Cohort Study). Typically, in a cohort study one or more generations are followed over time, that is, over their life-course. A cohort has been defined as ‘the aggregate of individuals who experienced the s...

Table of contents

• Cover Page
• Title Page
• Copyright Page
• Illustrations
• Examples
• Abbreviations
• Acknowledgements
• Introduction
• Part I: Longitudinal Research
• Part II: Longitudinal Analysis